Capacitor sound? As a first step in breaking new ground in relation to how capacitors can contribute to the ‘sound’ of a hi i amplifier, Cyril Bateman has designed a spot-frequency oscillator with sub-ppm distortion. any capacitors introduce distortions onto pure sinewave test signal. In some instances loading that the capacitor imposes on its valve or semiconductor driver. In others, the ‘capacitor generates the distortion within ise ‘Most properly designed power amplifiers measure less than 0.01% distortion when sinewave tested at IkHZ. This dis- tortion percentage equates fo 100 parts per milion, Such small distortions are believed to be inaudible, yet people ‘fen claim to hear distortions from these amplifiers when lis- tening to music “Many achorselaim to have identified differences in sound between different capacitor types. These differences have been ascertained not by measurements though, but by lis- tening tests, This has fed to a eetofit upgrade market sup- plying “better” audio grade capacitors at substantially ele- Vated prices compared to mass market types ‘A common subjectivist claim is tha oil-impregnated paper capacitors sound better than film types in valve amplifiers, Others claim that a PET capacitor sounds “why” while a polypropylene sounds ‘bright’, and that all cersmics soured ‘awful, Naturally these clanis have no supporting measure- ‘Many writers on this topic even decry measurements, pre- sumably in case such measurements disprove their subjec- tivist claims, Thave regularly rveived requests for advice about cap itors from readers who have read the many, oftea conflicting. subjectvist views about capacitor types. Over the year, these pages have also echoed to disputes between amplifier design- ‘ers and music enthusiasts regarding capacitor sound distor tion, These dispotes culminated in a particularly acrimonious debate a year Ago, during which I offered to perform some ‘comparative measurements ‘Asa long term capacitor designer and measurement engi- eet, [believe that any truly audible diferences must be both understandable and measurable. Understanding should be in terms of the capacitor constructions. Measurements may however require a change in measuring techniques. In order (0 develop suitable test methods, Lhave measured large numbers of capacitors of many types. From these mes- surements, [ have determiied the distortion differences between capacitor constrictions. ELECTRONICS WORLD July 2002‘What I did not expect to to fd ~ ant I find this rather is- turbing ~ is that within a small batch of capacitors, some ‘exhibit abnormally higher distortions, These anomalous capacitors typically exhibit some (en times greater istortion than others taped on the same card stip. In this, the first of a set of articles, { begin to honour my ‘commitment to quantify capacitor distortion. What the tests involved Using a scheme involving a test signal at kHz, itis possible to differentiate between capacitor types and between good or ‘bad capacitors within a type, Figs 1, 2. In all performance plots, the tkHz fundamental has been attenuated some 654B using a 1win-tee notch filter. The test capacitors for this article were each subjected toa three volts test signal, as measured across the capacitor terminals. Rather than perform measurements using. sophisticated ‘equipment, I decided to develop a low-cost method that could be easily replicated by any interested reader. In doing so, 1 hope to improve understanding of capacitors and reduce the umber of capacitor dispotes inthe letters pages. Initial investigations Spectrum analysers capable of measuring small distortion ‘components are prohibitively expensive for most peopic. ‘wanted to make sue tht performance measurements COU bbe made using readily available test gear like the Picoscope ADC-100 A-to-D converter or a compater sound card with FFT software I started by carrying out some initial eapacitor intermodu- lation tests, Experiments involving simple harmonic dis- tortion testing revealed easily interpreted differences when testing ess good capacitors. Testing good capacitors however ‘confirmed that my existing signal generators introduced far too much distonion. ‘A much better signal generator... Maving reviewed past fow-distorton oscillator designs, 1 brenl-burded the more promising ones. Usig these I tested 8 numberof capacitors but with only partial sucess From these results it became clear that 1 needed an extremely low dition kHz Snewave. had 0 Be able 1 “rive at least 3 voll into OSU near petect low ds tonion capacilive Yad, and doso without this load distorting ty test signal, Fig. 3. To test this near perfect capacitor, measured distortions of ny complete equipment needed to be less than ppm, oF 3.001%. This i approuching the order ofesillatr distor- tion produced by expensive measuring instruments such 38 those made by Audio Precision, ‘So bezan the design ofa suitable test oscillator with a price that woul be within the reach of most of you. The design of this oscillator forms the subject ofthis first article, Fig 4 Initial researches ‘My altention was caught by a remark about “fare Wien bridge oscillator desizn” in John Linsley Hood's 1981 description of 0.001% Wien bride osilatos, Most Wien bridge oscillators use a single amplifying stage. John sug- ested method spreading the capacitoriesistr elements Over two stages, This reduced the ive ao his ist amplifier and thus reduced its distortion, {ram some simulations shat sapponesJola's earlier views about lower distortion using this eoniguration. These simu lations also suggested a possible improvement. Usually the tno Wien bridge arms use equal value components. With John's new arrangement this sults in is cond amplifier having double the voltage output of his frst decide to double the capacitance ain halve the resiance ‘ofthe series combination. This would provide eal ontput COMPONENTS: Fig. 1. Some capacitors distort even a pure ThHz sinewave test signal. This Tonk X7R ceramic was made by a CECC approved, {European manufacturer twas tested at THz and 3 vol, in series with a 1012 current limiting resistor. Measurable dtstortion exist at all voltages down to 0.5 volts my lowest test Fig, 3. Plot ofa near perfect 1pFfoil/fim polypropylene ‘apacitor, tested at 3V in series with a 10082 current limiting resistor. lt clearly shows my target test specification has been attained. This excellent result depends as much on my output ampli desig as onthe oscillator. Combined distortions of my {est system and Ip load are busied inthe noise loor at 13048. July 2002 ELECTRONICS WORLD oFCOMPONENTS Fig 4. Final design forthe 1kHz test oscillator, with its ‘screening lid removed. Fitted with is lid but no other “shielding and with normal fluorescent room lighting it as used on my bench within 1 metre ofthe test PC forall ‘measurements. voltage from each amplifier with no change in oscillator Fe ‘quency. With two equal vedaze output stages, I could take the amplitude control voltage from one amplifier, leaving the ‘other able to provide my output signal. Trneeded 200,V drive into the negative inputs of both amplifiers to produce a 3 volts output, and this arrangement promised a high “Q° and low distortion. Many oscillators use a thermistor to controt oscillator amplitude. Distortion is then mostly third harmonic, which has been blamed on the thermistor, For my needs, third har- monic had io be minimised as much as possible. I needed a ifferent amplitude control ‘After some catalogue searching, I choose to design my amplitude control system around the Analogue Devices SSM2OI8P. This IC was expressly designed as a low-dis- tortion, audio-requency, vollage-controlled amplifier. Is Jowes distortion of 0.006% at IkHis produced with « 3 volt ‘input and OdB gain. For 0dB gain, a control voltage a few millivolts above OV is needed. Provided that this IC's output was used to supply only @ tiny portion of that drive needed to maintain osciliatinn its (0.006% distortion should contribute litle tothe circuit's out put. Thread-boarded the circuit using @ manual control voltage and with NESS34A ICs forthe oscillator. Encouraged by the eect ben ceed While I used ultra-low distortion, but expensive, AD797 ICs for U and Us ‘when building my final 1kHz oscillator, almost all its circuit development was done using low-cost NESS34 ICs, | found some 6d8 difference in distortion between these two IC types in my oscillator have ried other ICs for the oscillator, including the low-distortion OPAL34 and OPAG04. To facilitate evaluating ICs 1 Used Harwin turned pin sockets for each position. When using the AD797 for Uy and U2, itis preferable to fit a SOpF capacitor between pins 6 and 8. If you are using NE5534A ICs, itis preferable to fita 22pF capacitor instead between pins 5 and! 8 Neither capacitor is needed when using (OPAI34 or OPAGOS ICs. The oscillator tuning capacitors must be Jow.distortion types, preferably 1% extended foil with polystyrene, as shown in the photograph. However | have also built satisfactory working prototypes with 1% extended foif with polypropylene and 1% meaised polypropylene, n oder of preference. Obviously a good COG ceramic capacitor would work almost as well as ry first choice of polystyrene, provided the COG capacitor is available selected to Ifo tolerance. My PCB provides ‘mountings fora variety of suitable capacitors, The value of VR; needed to minimise distortion wil vary depending on which type IC and tuning capacitors are used. | found that only the NE55344 IC provided {ow distortion when used forthe output bor, Us, For this low/unity gain position, the 22pF capacitor is essential Aso for its gain contro, | found only one satisfactory variable resistor. That was @ Boums 91 series concluctive plastic, ‘bained as 148-557 from Farnell lar types may be OK too, but | have ret tried them, Don't use either cermet or Wirewound controls fr this position though. have tried several and they Certainly do not work acceptably The 50pF/22pF capacitors must be low- loss, low-distotion types. Polystyrene parts are preferable, but disc ceramics ~ COG only — can be used. Similarly for the remaining pcoarad capacors used, used COG ceramics for my prot ThePCB deawing provides or both altematives. In each case, my preferred IC choice is the first type listed on the schematic drawing. To produce such a low distortion oscillator its important to use resistors having a small voltage coefficient of resistance. To ensure an easily reproducible design, {used only 0.5% Welwyn RCS5C metal film resistors in the signal path. These are the black ‘components in the photograph. These are marke 2 on the seat These resistors use plated steel end caps, which | prefer for reliable long term ‘end contact stability. Many subjectivists claim non-magnetic end caps are better. do not subscribe to that belief. Undoubtedly, some of the oscillator ‘output distortion is generated inside the three mult-tun Cermet trimmers For wo positions, these trimmers are esse However the printed board does provi ‘mounting pads fora fixed resistor, which could be substituted for VR;, once its value has been determined during Calibration. a far Yhave retained use of the trimmer on my versions. While these RC55C types could be used throughout, for economy {used my standard, inexpensive 1% metal-ilm resistor, for all other positions Three bi-polar electrolytic capacitors are used in the gain control circuits These are the yellow-cased ‘Nita’ types Visible in the photograph. Equally suitable are the slightly langer Panasonic BP types. Both are stocked by Farnell, Do not use a conventional polar electrolytic capacitor {or these positions For such 2 low-distortion oscillator, itis essential to use good quality capacitors to clecouple the power supplies. For the 0.1 pF value, black in the photograph, | Used Evox-Rifa SMR, metallised yphenylene-suiphide film. | consider this film produces the best, small, low Cos, universal capacitor. They were obtained from RS, but unfortunately the ‘company has since stopped! supplying, them, ‘Alternatively, a good metallised PET itor, such as the Evox-Rifa MK or BC Components (Philips) 470 series, should be satisfactory. 1 used many of both these types, in my tand meter project. For the larger capacitors, | used BC Components” iF 470 series, grey in the photograph, and Rubycon YXF polar electrolytes. Again, other types should be ‘OK but they have not been tried in the circuit. In use the oscillator is powered from iy laboratory supply, set to output +18 volts. ELECTRONICS WORLD July 2002